CFeatureExtraction_LSD_BLD.cpp

Go to the documentation of this file.

/* +------------------------------------------------------------------------+
   |                     Mobile Robot Programming Toolkit (MRPT)            |
   |                          https://www.mrpt.org/                         |
   |                                                                        |
   | Copyright (c) 2005-2019, Individual contributors, see AUTHORS file     |
   | See: https://www.mrpt.org/Authors - All rights reserved.               |
   | Released under BSD License. See: https://www.mrpt.org/License          |
   +------------------------------------------------------------------------+ */

/*---------------------------------------------------------------
    CLASS: CFeatureExtraction
    FILE: CFeatureExtraction_LSD_BLD.cpp
    AUTHOR: Raghavender Sahdev <raghavendersahdev@gmail.com>
  ---------------------------------------------------------------*/

#include "vision-precomp.h"  // Precompiled headers

#include <mrpt/io/CMemoryStream.h>
#include <mrpt/otherlibs/do_opencv_includes.h>
#include <mrpt/system/os.h>
#include <mrpt/vision/CFeatureExtraction.h>  // important import

#ifdef HAVE_OPENCV_XFEATURES2D
#include <opencv2/xfeatures2d.hpp>
#endif
#ifdef HAVE_OPENCV_LINE_DESCRIPTOR
#include <opencv2/line_descriptor.hpp>
using namespace cv::line_descriptor;
#endif

using namespace mrpt::vision;
using namespace mrpt::img;
using namespace mrpt::math;
using namespace mrpt::img;
using namespace mrpt;
using namespace std;

#if defined(HAVE_OPENCV_XFEATURES2D) && defined(HAVE_OPENCV_LINE_DESCRIPTOR)
#define HAVE_OPENCV_WITH_LSD 1
#else
#define HAVE_OPENCV_WITH_LSD 0
#endif

void CFeatureExtraction::extractFeaturesLSD(
    const mrpt::img::CImage& inImg, CFeatureList& feats, unsigned int init_ID,
    unsigned int nDesiredFeatures, const TImageROI& ROI)
{
    MRPT_START

    mrpt::system::CTimeLoggerEntry tle(profiler, "extractFeaturesLSD");

#if (!HAVE_OPENCV_WITH_LSD)
    THROW_EXCEPTION(
        "This function requires OpenCV modules: xfeatures2d, line_descriptor");
#else
    using namespace cv;

    vector<KeyPoint> cv_feats;  // The opencv keypoint output vector
    vector<KeyLine> cv_line;

    // Make sure we operate on a gray-scale version of the image:
    const CImage inImg_gray(inImg, FAST_REF_OR_CONVERT_TO_GRAY);
    const Mat& theImg = inImg_gray.asCvMatRef();
    /* create a binary mask */
    cv::Mat mask = Mat::ones(theImg.size(), CV_8UC1);

    Ptr<LSDDetector> bd = LSDDetector::createLSDDetector();

    /* extract lines */
    cv::Mat output = theImg.clone();
    bd->detect(
        theImg, cv_line, options.LSDOptions.scale, options.LSDOptions.nOctaves,
        mask);

    // *All* the features have been extracted.
    const size_t N = cv_line.size();

    // Now:
    //  1) Sort them by "response":
    // sort the LSD features by line length
    for (size_t i = 0; i < N; i++)
    {
        for (size_t j = i + 1; j < N; j++)
        {
            if (cv_line.at(j).lineLength > cv_line.at(i).lineLength)
            {
                KeyLine temp_line = cv_line.at(i);
                cv_line.at(i) = cv_line.at(j);
                cv_line.at(j) = temp_line;
            }
        }
    }

    //  2) Filter by "min-distance" (in options.FASTOptions.min_distance)  //
    //  NOT REQUIRED FOR LSD Features
    //  3) Convert to MRPT CFeatureList format.
    // Steps 2 & 3 are done together in the while() below.
    // The "min-distance" filter is done by means of a 2D binary matrix where
    // each cell is marked when one
    // feature falls within it. This is not exactly the same than a pure
    // "min-distance" but is pretty close
    // and for large numbers of features is much faster than brute force search
    // of kd-trees.
    // (An intermediate approach would be the creation of a mask image updated
    // for each accepted feature, etc.)

    unsigned int nMax =
        (nDesiredFeatures != 0 && N > nDesiredFeatures) ? nDesiredFeatures : N;
    const int offset = (int)this->options.patchSize / 2 + 1;
    const size_t size_2 = options.patchSize / 2;
    const size_t imgH = inImg.getHeight();
    const size_t imgW = inImg.getWidth();
    unsigned int i = 0;
    unsigned int cont = 0;
    TFeatureID nextID = init_ID;

    if (!options.addNewFeatures) feats.clear();

    /* draw lines extracted from octave 0 */
    if (output.channels() == 1) cvtColor(output, output, COLOR_GRAY2BGR);

    while (cont != nMax && i != N)
    {
        KeyLine kl = cv_line[i];
        KeyPoint kp;

        if (kl.octave == 0)
        {
            Point pt1 = Point2f(kl.startPointX, kl.startPointY);
            Point pt2 = Point2f(kl.endPointX, kl.endPointY);

            kp.pt.x = (pt1.x + pt2.x) / 2;
            kp.pt.y = (pt1.y + pt2.y) / 2;
            i++;
            // Patch out of the image??
            const int xBorderInf = (int)floor(kp.pt.x - size_2);
            const int xBorderSup = (int)floor(kp.pt.x + size_2);
            const int yBorderInf = (int)floor(kp.pt.y - size_2);
            const int yBorderSup = (int)floor(kp.pt.y + size_2);

            if (!(xBorderSup < (int)imgW && xBorderInf > 0 &&
                  yBorderSup < (int)imgH && yBorderInf > 0))
                continue;  // nope, skip.

            // All tests passed: add new feature:
            CFeature ft;
            ft.type = featLSD;
            ft.keypoint.ID = nextID++;
            ft.keypoint.pt.x = kp.pt.x;
            ft.keypoint.pt.y = kp.pt.y;
            ft.x2[0] = pt1.x;
            ft.x2[1] = pt2.x;
            ft.y2[0] = pt1.y;
            ft.y2[1] = pt2.y;
            ft.keypoint.response = kl.response;
            ft.keypoint.octave = kl.octave;

            if (options.patchSize > 0)
            {
                mrpt::img::CImage p;
                inImg.extract_patch(
                    p, round(kp.pt.x) - offset, round(kp.pt.y) - offset,
                    options.patchSize,
                    options.patchSize);  // Image patch surronding the feature
                ft.patch = std::move(p);
            }
            feats.emplace_back(std::move(ft));
        }
        ++cont;
        // cout << ft->x << "  " << ft->y << endl;
    }

#endif
    MRPT_END
}

void CFeatureExtraction::internal_computeBLDLineDescriptors(
    const mrpt::img::CImage& in_img, CFeatureList& in_features)
{
#if (!HAVE_OPENCV_WITH_LSD)
    THROW_EXCEPTION(
        "This function requires OpenCV modules: xfeatures2d, line_descriptor");
#else

    mrpt::system::CTimeLoggerEntry tle(
        profiler, "internal_computeBLDLineDescriptors");

    using namespace cv;

    if (in_features.empty()) return;

    const CImage img_grayscale(in_img, FAST_REF_OR_CONVERT_TO_GRAY);
    const Mat& img = img_grayscale.asCvMatRef();

    vector<KeyPoint> cv_feats;  // OpenCV keypoint output vector
    Mat cv_descs;  // OpenCV descriptor output

    cv::Mat mask = Mat::ones(img.size(), CV_8UC1);

    BinaryDescriptor::Params params;
    params.ksize_ = options.BLDOptions.ksize_;
    params.reductionRatio = options.BLDOptions.reductionRatio;
    params.numOfOctave_ = options.BLDOptions.numOfOctave;
    params.widthOfBand_ = options.BLDOptions.widthOfBand;

    Ptr<BinaryDescriptor> bd2 =
        BinaryDescriptor::createBinaryDescriptor(params);
    /* compute lines */
    std::vector<KeyLine> keylines;

    bd2->detect(img, keylines, mask);

    /* compute descriptors */
    bd2->compute(img, keylines, cv_descs);
    keylines.resize(in_features.size());

    // -----------------------------------------------------------------
    // MRPT Wrapping
    // -----------------------------------------------------------------
    int i = 0;
    for (auto& ft : in_features)
    {
        // Get the BLD descriptor
        ft.descriptors.BLD.emplace();
        auto& desc = ft.descriptors.BLD.value();
        desc.resize(cv_descs.cols);
        for (int m = 0; m < cv_descs.cols; ++m)
            desc[m] = cv_descs.at<int>(i, m);
    }

#endif  // end of opencv3 version check
}  // end internal_computeBLDDescriptors